Re: [open-cobol-list] Preliminary results on gdb support

[Jeff Chimene]

On 06/09/2009 05:10 PM, Jeff Chimene wrote:
> Hi,
>
> I have some preliminary results on adding debugging support to OC.
>
> I'd like to get some feedback from the list as to whether it's worth 
> pursuing.
>
> This work is based on OC v 1.0. If it's worth continuing, it can't be 
> too difficult to port to 1.1
>
> Consider the following COBOL program:
>        PROGRAM-ID.      prog.
>        DATA             DIVISION.
>        WORKING-STORAGE  SECTION.
>        01 G             VALUE "012345678".
>          02 X           PIC X OCCURS 4.
>          02 Z           PIC X OCCURS 1 TO 4.
>        01 H                PIC 99.
>          88 myriad        value 1, 3, 9 thru 16, 17, 25 thru 50.
>        PROCEDURE        DIVISION.
>            DISPLAY X((3 + 1) / 2) NO ADVANCING.
>            DISPLAY X(2 ** 2) NO ADVANCING.
>            DISPLAY Z(1).
>
>            STOP RUN.
>
> Here is a sample run using a simple gdb user-defined command "view" 
> (See below)
>
> $ gdb hello
> Breakpoint 1 at 0x8048b5f: file /home/jchimene/hello.cob, line 11. 
> <=== This breakpoint is set in a project-specific .gdbinit: "break 
> hello.cob:11"
> (gdb) run
>
> Breakpoint 1, prog_ (entry=0) at /home/jchimene/hello.cob:11
> 11                 DISPLAY X((3 + 1) / 2) NO ADVANCING.

Change the following "X" to "G"

> (gdb) view "X" <=== Notice the quotation marks
> $1 = (unsigned char *) 0x8049360 "01234567"
>
> (gdb) view "H"
> $2 = (unsigned char *) 0x8049350 "00"
>
> (gdb) quit
>
> The simple way to view COBOL data items is based on the following:
>
> Modify cobc to take the information from the CB_FIELD structure and 
> convert it to the following structure:
>
>    static struct symtab {
>         char const         *cobname;
>         unsigned char    *cname;
>    }
>
> This structure is defined and initialized in the program's .h file 
> (e.g. hello.c.h)
>
>    static struct symtab {
>         char const         *cobname;
>         unsigned char    *cname;
>    } st[] = { { "X", b_5 }, {"H", b_8}, {'\0'} }

Change the "X" above to "G"

> The scope of the variables defined in the .h file is the 
> compiler-generated routine "prog_()". This is rather mysterious; 
> there's no good way to /a priori/ set COBOL-specific breakpoints [*]. 
> For now, it looks like an inspection of the C source combined with a 
> project-specific .gdbinit is the solution.
>
> ----------
> [*]
>
>    i.e. a BREAK MAIN produces no useful work as this is the runtime
>    entry point; which is not the same as the start of the procedure
>    division. There doesn't seem to be a good way to always identify the
>    start of the procedure division. I'm probably missing something 
> obvious.
>
> The next modification involves a gdb initialization file that searches 
> the structure on COBNAME then prints the corresponding CNAME. This 
> user-defined command is defined in a project-independent 
> $HOME/.gdbinit (or wherever it is for Windows users)
>
>    define view
>         set $r0 = 0
>         set $r1 = st[$r0].cobname
>         while ($r1 > 0)
>             if (0 == strcmp($r1,$arg0))
>                 print st[$r0].cname
>             end
>             set $r0 = $r0 + 1
>             set $r1 = st[$r0].cobname
>         end
>
> The big issue right now is to extend the SYMTAB structure 
> initialization to handle the various ways that cobc implements (record 
> definition vs. working storage vs. constants) and declares COBOL data 
> items (integer, float, &c). The controlling logic is fairly complex, 
> and I don't know if duplicating it for the purposes of creating the 
> SYMTAB structure will be a useful task.

Another, big(ger) issue is to add support for child fields.